Prevent anti-aliasing for imshow in matplotlib

Question

When I use matplotlib's imshow() method to represent a small numpy matrix, it ends up doing some smoothing between pixels. Is there any way to disables this? It makes my figure's misleading in presentations.

The figure above is a 28x28 image, so I should be seeing large squares of single colors representing each pixel (as matlab would display it when using imagesc()). But Instead, the pixels seem to be blurred with neighboring pixels. Is there a way to disable this behavior?

Solution 1

There is an interpolation option for imshow which controls how and if interpolation will be applied to the rendering of the matrix. If you try

imshow(array, interpolation="nearest") 

you might get something more like you want. As an example

A=10*np.eye(10) + np.random.rand(100).reshape(10,10)
imshow(A)

compared with

A=10*np.eye(10) + np.random.rand(100).reshape(10,10)
imshow(A, interpolation="nearest")

Solution 2

you can also try the function

matshow 

which name indicated that it does exactly what you asked - represent matrices. It is quite handy when you do not need to customise the figure too much.

BTW, one of the best resources for matplotlib is their Gallery

Comments

[Christopher Dorian]

Thanks, that seems to work. Out of curiosity, why is it called 'nearest' for interpolation? The documentation on this page: matplotlib.sourceforge.net/api/… doesn't describe what 'nearest' does.

[talonmies]

To be honest, I don't know. In the development tree there is now also an interpolation="none" option which does pretty much the same thing. This example provides a little more information on what interpolation="nearest"is trying to achieve.

[DrSAR]

it is maybe trying to sample from the 'nearest' neighbour? in the vicinity of the display pixel

[wim]

it is indeed 'nearest neighbour' interpolation

[Ander Biguri]

@ChristopherDorian , just comming here 4 years late, but the difference between nearest and none: matplotlib.org/examples/images_contours_and_fields/…